Imports for scikit-learn 1.7 Release Highlights: Improved HTML Displays, Custom Validation Sets, and ROC from CV Results¶
This release enhanced estimator HTML representations with parameter tables (highlighting non-default values in orange and providing copy buttons for parameter names), added custom validation set support for HistGradientBoosting estimators, and introduced RocCurveDisplay.from_cv_results for plotting ROC curves from cross-validation: The custom validation set feature allows passing X_val and y_val directly to HistGradientBoostingClassifier.fit, which is used for early stopping evaluation instead of splitting the training data internally. Combined with Pipeline’s transform_input=["X_val"], validation data flows through the same preprocessing transformers as training data, ensuring consistent feature engineering.
Additional highlights include Poisson loss support and sample weight handling for MLPRegressor/MLPClassifier, SciPy sparse array compatibility across all estimators, and expanded Array API support for GPU computation: RocCurveDisplay.from_cv_results accepts the output of cross_validate(return_estimator=True, return_indices=True) and plots individual fold ROC curves on the same axes, providing a visual assessment of model stability across cross-validation splits. The migration from SciPy sparse matrices to sparse arrays prepares scikit-learn for the eventual deprecation of the matrix interface, ensuring forward compatibility with future SciPy versions.
# ruff: noqa: CPY001
"""
=======================================
Release Highlights for scikit-learn 1.7
=======================================
.. currentmodule:: sklearn
We are pleased to announce the release of scikit-learn 1.7! Many bug fixes
and improvements were added, as well as some key new features. Below we
detail the highlights of this release. **For an exhaustive list of
all the changes**, please refer to the :ref:`release notes <release_notes_1_7>`.
To install the latest version (with pip)::
pip install --upgrade scikit-learn
or with conda::
conda install -c conda-forge scikit-learn
"""
# %%
# Improved estimator's HTML representation
# ----------------------------------------
# The HTML representation of estimators now includes a section containing the list of
# parameters and their values. Non-default parameters are highlighted in orange. A copy
# button is also available to copy the "fully-qualified" parameter name without the
# need to call the `get_params` method. It is particularly useful when defining a
# parameter grid for a grid-search or a randomized-search with a complex pipeline.
#
# See the example below and click on the different estimator's blocks to see the
# improved HTML representation.
from sklearn.linear_model import LogisticRegression
from sklearn.pipeline import make_pipeline
from sklearn.preprocessing import StandardScaler
model = make_pipeline(StandardScaler(with_std=False), LogisticRegression(C=2.0))
model
# %%
# Custom validation set for histogram-based Gradient Boosting estimators
# ----------------------------------------------------------------------
# The :class:`ensemble.HistGradientBoostingClassifier` and
# :class:`ensemble.HistGradientBoostingRegressor` now support directly passing a custom
# validation set for early stopping to the `fit` method, using the `X_val`, `y_val`, and
# `sample_weight_val` parameters.
# In a :class:`pipeline.Pipeline`, the validation set `X_val` can be transformed along
# with `X` using the `transform_input` parameter.
import sklearn
from sklearn.datasets import make_classification
from sklearn.ensemble import HistGradientBoostingClassifier
from sklearn.model_selection import train_test_split
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
sklearn.set_config(enable_metadata_routing=True)
X, y = make_classification(random_state=0)
X_train, X_val, y_train, y_val = train_test_split(X, y, test_size=0.2, random_state=0)
clf = HistGradientBoostingClassifier()
clf.set_fit_request(X_val=True, y_val=True)
model = Pipeline([("sc", StandardScaler()), ("clf", clf)], transform_input=["X_val"])
model.fit(X, y, X_val=X_val, y_val=y_val)
# %%
# Plotting ROC curves from cross-validation results
# -------------------------------------------------
# The class :class:`metrics.RocCurveDisplay` has a new class method `from_cv_results`
# that allows to easily plot multiple ROC curves from the results of
# :func:`model_selection.cross_validate`.
from sklearn.datasets import make_classification
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import RocCurveDisplay
from sklearn.model_selection import cross_validate
X, y = make_classification(n_samples=150, random_state=0)
clf = LogisticRegression(random_state=0)
cv_results = cross_validate(clf, X, y, cv=5, return_estimator=True, return_indices=True)
_ = RocCurveDisplay.from_cv_results(cv_results, X, y)
# %%
# Array API support
# -----------------
# Several functions have been updated to support array API compatible inputs since
# version 1.6, especially metrics from the :mod:`sklearn.metrics` module.
#
# In addition, it is no longer required to install the `array-api-compat` package to use
# the experimental array API support in scikit-learn.
#
# Please refer to the :ref:`array API support<array_api>` page for instructions to use
# scikit-learn with array API compatible libraries such as PyTorch or CuPy.
# %%
# Improved API consistency of Multi-layer Perceptron
# --------------------------------------------------
# The :class:`neural_network.MLPRegressor` has a new parameter `loss` and now supports
# the "poisson" loss in addition to the default "squared_error" loss.
# Moreover, the :class:`neural_network.MLPClassifier` and
# :class:`neural_network.MLPRegressor` estimators now support sample weights.
# These improvements have been made to improve the consistency of these estimators
# with regard to the other estimators in scikit-learn.
# %%
# Migration toward sparse arrays
# ------------------------------
# In order to prepare `SciPy migration from sparse matrices to sparse arrays <https://docs.scipy.org/doc/scipy/reference/sparse.migration_to_sparray.html>`_,
# all scikit-learn estimators that accept sparse matrices as input now also accept
# sparse arrays.